An Encapsulation for Reasoning, Learning, Knowledge Representation, and
Reconfiguration Cognltive Radio Elements
Keith E. Nolan *
CTVR
Trinity College Dublin
[email protected]
Paul Sutton
CTVR
Trinity College Dublin
[email protected]
Abstract
State and contextual awareness, reasoning and conclusions formation, and a means of directing application,
structural and parameter-level radio reconfiguration are
key elements of a cognitive radio. This paper describes
a cognitive radio design capable of scaling between the
two extremes of minimal cognitive capabilities and complex highly-evolved cognitive radio abilities, which is being adopted for real tests using licensed cognitive radio
test spectrum. A memory element stores state, sensor, objectives, actions and conclusions information and the relevance of this information can be varied in order to identify or ignore common traits or occurrences. The decisionmaking and conclusionsformation abilities of this cognitive
radio design can use (or choose to ignore using the variable weightingfacility) external information relating to the
network and etiquettes in conjunction with the memory eement. A set ofactionsformulated by the reasoning and conclusions formation stages direct the radio reconfiguration.
This design is implemented using a General-Purpose Processor (GPP) platform as it currently offers the very high
level ofreconfigurability requiredfor very malleable cognitive radio design.
1. Introduction
This section introduces the idea of cognition and identifies the core requirements for a cognitive radio.
Cognition in signal-processing and system control terms
is the ability to develop contextual and, environmental
awareness aiding the development of an optimal solution
for a particular problem, recognise developing patterns of
Scienlce
* This material is based upon work supported by
Fourndatioun
IrelanLd unlder Gralnt No. 03/CE3/I405 as par of the Celntre for Telecommunlications Vallue-Chainl Research (CTVR) at Trinlity Colllege Dubllin, tre-
lalnd.
Linda E. Doyle
CTVR
Trinity College Dublin
ledoylegtcd.ie
behavior, respond to the time-varying nature of wireless
channel and user activity and learn from previous experiences. Considered in isolation, the foundations of each of
the core observe, orient, react and learn stages of the cognition cycle first described by Mitola L] and more recently
by Haykin [2], are not new concepts. However, it is the
innovative application of a combination ofthese techniques
in a cognitive wireless communications context that is innovative. Observation information can be derived from internal radio and system activity (including available resources,
radio capabilities and spectrum activity detected at the receiver), and external sources (including external environmental sensors, policies, network-level information).
An implementation of a cognitive enti requires a
highly-reconfigurable core, which can change and evolve
according to the orient, react and learn stages in the cogni-
tive cycle. Popular approaches taken in relation to how and
why changes are necessary are based on game-theoretic [5],
genetic algorithmic [6], Fuzzy Logic [8] and artificial
neural-network principles [7]. These techniques are used
to find. an optimum (or near-optimum) solution to a paicular wireless communications problem but require a reconfigurable radio in order to implement the desired changes
and analysis the implications of this change. It is feasible
that fully-engaged cognitive abilities are always required
depending on the complexity of, and challenges presented
by particular scenarios. Therefore, the ability to change between minimal and complex cognitive behavior can potentially reduce power consumption and increase the operating
lifetime of the device.
Section 2 describes the reconfigurable core, which is the
key enabling feature for this cognitive wrapper, Section 3 is
ainldn
description
of the coreprcsig
wrapper
elemenats of the coglnitive
th awrns
ato
man
le.nn
tiation stages. In SectionL 4 presenLts solme conclusionks frolm
this work.
=F
1-4244-03891
-2106$00 ©20 IEEE
Authorized licensed use limited to: TRINITY COLLEGE DUBLIN. Downloaded on May 6, 2009 at 06:06 from IEEE Xplore. Restrictions apply.
2, Reconfigurable Radio
This section briefly describes the term reconfigurable radio and one instance of an actual system that is used as the
basis for the cognitive wrapper described in this paper.
The term reconfigurable radio is used in this paper to
describe a heteromorphic radio signal-processing chain implemented in software, connected to a minimal hardware RF
front-end that may itself be reconfigurable through physical
change or under software control. A reference design and,
implementation of a reconfigurable radio used as the basis
of the reasoning wrapper design in this paper is called Implementing Radio In Software (IRIS) [9]. This system uses
eXtensible Markup Language (XML) [3] to describe a radio
in terms of a signal-processing chain of elements called Radio Components. Examples of existing Radio Components
include modulators, demodulators, access schemes, filters,
signal conversion, source and sink elements. These Radio
Components can either be sourced. from a local inventory
of available Components (created by the designers) or from
one or more remotely-located inventories using a wired link.
Each of these elements has a common architectural framewor k facilitating rapid development and straightforward internal creation, execution and tear-down processes.
The IRIS system caters for a hierarchy of possible reconfiguration tasks called action sets. These action sets are developed, by the observations reporting, awareness processing and reasoning engine loop as depicted by Fig. 1. Application reconfiguration allows the replacement of an entire signal-chain with another desired signal chain in order to change the active application. Component reconfiguration enables one or more signal-chain processing elements (Radio Components) to be removed/replaced/added
at will. This reconfiguration can during run-time in addition to the trivial static-case reconfiguration scenario. Dynamic parameter-level reconfiguration is also possible and
all relevant parameters used in each Radio Component can
be changed on demand. These reconfiguration possibilities
allow the radio core to be molded into any form according to
the instructions of a higher-level entity (cognitive wrapper),
which in turn is possibly in response one or more reconfiguration triggers or drivers.
The higher-level IRIS entity governing change within
the reconfigurable core is called Control Logic [9].
This is a software mechanism that implements the replace/add/remove Radio Components and controls the cascade of reconfiguration required when parameters are
changed within one or more Radio Components that may
impact on other fRadio Components further alonlg the signalchain. This Control Logic has been expanded. to cater for a
reasoning engine, :me:mory delLay-lLine and external input intelrfaces as described in this pape:r.
2.1. Reconfiguration Drivers
Preparing for the possibility of change within a cognitive radio does not imply that continuous change is required,
during the operating lifetime of the radio. It is conceivable
that a static architecture is sufficient in some cases. Reconfiguration activity is triggered when the cognition engine
determines that an observed event(s) or states necessitate an
application, component or parameter change. It is necessary
to drive these reconfiguration processes in an intelligent
manner that will result in the implementation of a desired
preset feasible solution or a solution developed. by the reasoning and conclusions formation engine. Reconfiguration
drivers are not limited to internal device events/observations
however, and can account for radio, network, regulatory
and physical environment changes, and application, business and social context changes.
Observations
Awareness
Processing
&
Reconfigurable
Core
Action Set
Reasoning
Figure 1. Reconfigurable core showing inputs (set of actions) and outputs (device
state, capabilities and spectrum observa-
tions)
3. Cognitive Wrapper
In this section, the primar contribution of this paper
is presented in more detail. This is a realisable cognitive
wrapper with scalable-'intelligence' and designer-specified
learning and reasoning algorithm capabilities. This section
describes the core entities comprising this cognitive wrapper design, where the key fundamentals of the design are
shown in Fig. 2.
The cognitive wrapper described in this paper encapsulates a reconfigurable core, which in this case is the IRIS
system. Features of this wrapper, as illustrated in Fig. 2
include the observation, awareness and knowledge representation mechanisms, a variable-length memory delayline used to store current and historical knowledge sets,
which can also be used, to identify (or disregard) common
traits/characteristics. The reasoninag engilne generates the reconfiguration tasks and d.irects these changes in the reconfigurable core. This diagram also ilUlustrates that constraints
which can include etiquettes fo:r radio-behavior, a lmeasure
Authorized licensed use limited to: TRINITY COLLEGE DUBLIN. Downloaded on May 6, 2009 at 06:06 from IEEE Xplore. Restrictions apply.
of the system capabilities and regulatory policies can also
have a direct influence on the reasoning and conclusionsformation processes.
Reasoning tasks include developing the sequence of application, structural and parameter changes (action sets) or
deciding that no reconfiguration is necessary. We consider a
full-featured highly-involved, cognitive radio device for the
following descriptions of the reasoning wrapper capabilities.
Implementation of a cognitive system requires
awareness-formation, reasoning and learning, and,
A cognitive
conclusions-development capabilities.
radio therefore requires a means of observing the environmental, social, user, spectrum and policy landscapes
as described in Section 2.1, memorising (or choosing not
to remember) previous events, actions and consequences,
decision-making and conclusions-formation. The ability
to mould the reconfigurable core by executing actions that
direct the operation and structure of this core is also a high
priority objective. For maximum system flexibility, the
radio device should have the ability to scale the influence
of the cognition capabilities between the two extremes of a
highly-involved cognitive radio to a basic device with no
cognition capabilities.
Reasoning Engine
Knowledge Representation Delay-Line
i
Memory{Tasks:Actions:Outcomes: Conclusions}
m
Obevton
O1
*s NN
Observationims
O X 2z 3s .............
Long Term
lfStotTeem
Reconfi
fgurabl
Core
. oiguVabie
li
Actionl ket
Vaibl
Decision-Making, Learning,
Conclusionls Formation
ConclusionsFo
Action
et
I____
I
0
Variable
e Weighting:
egulatory
System
Constraints
n
0
Corintst§
Policy
Figure 2. Reasoning wrapper overview illustrating the knowledge representation delayline, reasoning and learning engine, constraints, and reconfigurable core entity
A highly-evolved cognitive radio can employ contextual
reasoning to help determine the best course of action to
take. Interpretation of selected internal and external physi-
cal, spatial, environmental, political and objectives is therefore necessa to develop and maintain contextual awareness during the lifetime ofthe cognitive radio. O'bservations
may originate from internal and externalL sources. The set of
internal source informration inclLudes availLalble energy, exist-
ing components, data-type descriptions, available processing power, RF front-end capabilities, networking capabilities and, fixed, mobile, nomadic mobility status information.
Extra sensing information can be obtained, from environmental, spatial and biometric sensors including temperature, pressure, air and water quality, shock and vibration
information. Spatial awareness is not limited to geographical location but include trajectories, altitude and device-tilt
information. Awareness of the time-value of this information is a critical factor in the cognitive control mechanisms.
Available spectrum may have a finite usage window, reaction to a sudden shock experienced by the cognitive radio
may require immediate countermeasures and a device faced
with a dwindling energy supply may have to initiate graceful degradation or backup measures before the remaining
energy is depleted fully. Instead of the power-inefficient
case where all possible sensing sources are activated, at all
times, the cognitive radio must be capable of focusing its
resources of sensing sources deemed important at any particular time and deactivate sensing sources considered irrelevant.
3.1. Knowledge Representation
of the radio, including current and previous rastates, radio resources,
and internal and external obserdioAspects
vation are knowledge sources used as part of the radio cognition processes. It is the relevance of this infoirmation in
a particular context, instance or period of time, or scenario
that influences the value of this knowledge however. An
to store the sequence of actions taken and measurable consequences of these actions is therefore a valuable
asset. Information derived from some source entity often
has a strict description syntax. In order to interpret this information correctly therefore, devices must conform to a
common syntactical convention. XML (eXtensible Markup
Language) for example, is a portable method of representing information, which can be parsed by software processes
and is presented in a human-interpretable form [3]. Web
Ontology Language (OWL) is a method of representing information that does not necessarily have to be presented in a
human-readable form but this information is essentially derived from an English language description of the scenario
or task [4]. OWL offers a means of specifying the semantics of a scenario, which can be conveyed and translated by
platforms with different syntax conventions.
The ability to store, order, extract and reuse information relating to current and historical state, actions, conclusions, objectives in a structured format facilitates application of this information in the cognitive decision-making
processes. Ultimately this enalbles the cognitive radio to
make lbetter operationalL decisions. KnowlLedge of previous actions, and consequences of these actions also aids the
ability
Authorized licensed use limited to: TRINITY COLLEGE DUBLIN. Downloaded on May 6, 2009 at 06:06 from IEEE Xplore. Restrictions apply.
ficial neural networks, Bayesian or Fuzzy system logic imforward-planning and anticipative action ofthe cognitive raplementation approach. This is achieved using the Control
dio. Emerging problems can be decomposed into a set of
Logic interface that provides the means by which, external
problems with less complexity. Solutions to these nested
processes can attach to, and direct the reconfigurable core.
challenges may already exist within the stored knowledge
This stage can also be de-activated using this Control Logic
sets thus potentially reducing the overall solution-formation
interface if a minimal-cognition or non-cognitive device optime.
eration is required.
The memory delay-line shown in Fig. 2 is the means used
The feasibility of a decision making, learning and conto store current and historical knowledge sets for the cognitive radio system described in this paper. A method used to
clusions formation approach is dependent on the time required to present viable solutions and the implementation
represent short and long-term knowledge, which forms part
ofthe input and ultimately influences the reasoning wrapper
complexity associated with each approach. The presentation of a solution approaching optimality within the time
outputs and desired actions. Analogous to a finite-length fi'ter delay-line which stores current and historical knowledge
constraints allowed has a potentially greater value than an
sets. Information from all stored memory sets is available
optimal solution that is produced too late i.e. after the imfor use by the cognitive engine.
plementation deadline. Complexity and the processing burden can be reduced by implementing some features of a
The relevance of certain aspects of each knowledge set
stored in the memory delay-line may not be constant. A
chosen approach. It is conceivable that significant gains using dynamic spectrum access techniques can be achieved
memory-merging capability offers some interesting possibilities. For some scenarios, identification of common
without the full weight of a maximal-complexity cognitive
engine. The platform presented in this paper offers the
traits, actions or consequences of previous radio reconfiguration and observed events may be more important than
ability to investigate the real achievable spectral-efficiency
gains using actual RF spectrum in a controlled interference
spurious events or actions. Selective memory can also be
used to place a greater bias on recent knowledge rather than
and user-activity environment. The cost function determining the real increase in spectrum-usage efficiency can be
longer-term knowledge, or vice-versa. Weighting factors,
reconfigured permitting the exploration of many different
analogous to filter coefficients are used to implement memcase studies. Examples of these include investigating the
ory selectiveness. The selective nature is reconfigurable by
cost of rapid spectrum allocation where processing power
varying the weighting factors associated with the knowlis the determining factor and investigating the cost of opedge set stored in each memory delay. Equation 1 is a
conceptual example of this process where y(k) represents
portunistic access of narrow spectrum segments where inthe kth desired parameter value/solution, v(n) is the nrh
terference may be the important factor.
weighting factor assigned by the reasoning engine, x(n) is
A cognitive radio faced with a developing wireless comm
b f
t e
c
the nth knowledge set element stored in the memory delaymunications sscenario m
may be forced to expend considerable
line and
is the memory-length. The twoextreme
two extreme...
lineNdelays
and Ndy,isthememory-length.
energy using resources on determining the best course of
cases of 1. a memory-less radio device is achievable by
action. A better approach is to break down the developing
assigning a weighting factors of zero for all memory delaysituation and apply a sequence of less complex incremental
line weights and 2. aphotographic memory is achievable by
solutions. The objective in this case is to solve the complex
assigning a weighting factor of one for all memory delayoverall problem using a combination of these incremental
line weights and averaging the result.
solutions. The possibly complex scenario can decomposed
into
two main classes, where incremental solutions for each
Ndelays
(1)
E w(T)sr(n)
stage may already exist in the delay-line of knowledge sets.
y(k)
The first case is a repeatable scenario and the second case
n=O
is a unique scenario.
3.2. Decision Makling, Learning and ConA repeatable scenario is where similar wireless comciuslons Format'ion
munications tasks and observable environmental conditions
occur more than once. In this case, the cognitive radio,
The main objective of the decision making, learning and
which identifies the emerging similarities from the knowlconclusions formation element of the reasoning wrapper is
edge sets, can invoke a sequence of previously successful
to produce an 'intelligent' and timely answer to a problem
procedures in an attempt to accelerate the completion of
set based on previous actions and consequences, current obcommunications task(s). The potential benefits of this abilservations and o'bjectives and descriptions of the data- ples
ity include conservationl of radio resources, increased userused for the desired wire'less communications task.
satisfaction levels and possi'bly reduced interference levels
This cognitive stage, ilUlustrated in Fig. 2 is designed to
to other wireless devices as the signalUling-overheads mray
adopt any feasible galme-theoretic, genetic algorithmic, artibe reduced as a consequence.
Authorized licensed use limited to: TRINITY COLLEGE DUBLIN. Downloaded on May 6, 2009 at 06:06 from IEEE Xplore. Restrictions apply.
Table 1. Knowledge Set Example
fJ
Observations
Tasks
BW: 2MHz Freq: 2.08GHz
Voice: High quality
Avail. eaergy:high
Reconfigure: OFDM
Users: 2
Maintain link
Mobility: mobile
The unique scenario is where internal and external ob-
servations, available radio resources and desired communications task are not the same (or in the same sequence) as
previously experienced by the cognitive radio. It is possible
that the unique scenario may be decomposed as a sequence
of repeatable scenarios. Thus the cognition wrapper can attempt a combination of proven tactics in order to provide a
solution.
The knowledge gained from a communications task may
include the data-type involved, power and duration of the
transmission, structural, component and parameter configuration of the radio that successfully (or unsuccessfully)
completed the task. The consequences of this communications task can include Bit Error Rates (BER), Quality of
Service (QoS), consumed radio resources, interference experienced or inflicted, and an estimate of the spectrum efficiency.
An action set is the list of required components, structural configurations or possible reconfiguration instructions
for existing structures, component parameter-values and
deadlines by which the action set should be implemented.
Actions
OFDM: 500 user share
TX Power: mm.
Sense before use
El
Policies
Interference Avoidance
Social: extrovert
References
[1]
Mitola, J., III; Maguire, G.Q., Jr., "Cognitive ra-
dio: making software radios more personal," Personal
Communications, IEEE [see also IEEE Wireless Communications], vol.6, no.4pp. 13-18, Aug 1999
[2] Hlaykin S., Cognitive Radio: Brain-Empoweredd
Wireless Communications, IEEE Journal on Selected
Areas in Communications, February, 2005.
[3]
http://www.w3.org/XML/
[4] http://www.w3.org/TR/owl-features/
[5] J. Neel, J. Reed, R. Gilles. The Role of Game Theory in the Analysis of Software Radio Networks, SDR
Forum Technical Conference November, 2002.
[6] Rieser, CJ; Rondeau,
Bostian, C.W; Gall
land Testi "oa ribt teneti orithm baise
T.W.;
g
gg
aepouebycognitive engine for programmable radios," Military
Communications Conference, 2004. MILCOM 2004.
bythe ognitie
and leai
rening
mechansms
emloyed n the
engie.
mechansesms
employed
vol.3, no.pp. 1L437- 1443 Vol. 3, 31L Oct.-3 Nov.
Thehet-IEFF,
2004
eromorphic reconfigurable radio core is instructed to exeAction
sets
cute an action set using Control Logic. Control Logic is
the means of addressing each structure, component and parameter capable of being implemented by the reconfigurable
core. The action set for an OFDM receiver can include the
entire set of required components, or incremental changes
to an existing structure, required parameter values, and the
time-span in which to implement the action set.
[7] Hopfield, J.J., "Artificial neural networks," Circuits
andDevices Magazine, IEEE, vol.4, no.5pp.3- 10, Sep
1988
[8] Klir, G.J., "Fuzzy logic," Potentials, IEEE , vol. 14,
no.4pp. 10-1L5, Oct/Nov 1995
[9] P. Mackenzie, "Reconfigurable Software Radio Sys4. Conclusions:tems", Ph.D dissertation, Trini College Dublin, Ireland, 2004.
This paper has presented a cognitive wrapper encapsulat[10] Nolan, K.E., Reconfigurable OFDM Systems, Ph.D
ing a highly reconfigurable radio core. The influence of this
thesis, University of Dublin, Trinity College, May
wrapper can be varied in order to vary the 'intelligence' of
2005.
the cognitive radio between the two extremes of a baseline
radio, and a highly-complex and evolved system to account
for scenarios where complex cognition may not be necessa O'bservations, actions and conclusions developedby
the cognitive radio can be stored in a variable-length memoydelay-lLine. A selective mrelmo mechanis elnable the
relvac of aset of thi inomai ton bevaried.
Authorized licensed use limited to: TRINITY COLLEGE DUBLIN. Downloaded on May 6, 2009 at 06:06 from IEEE Xplore. Restrictions apply.